I just saw NVIDIA offers this product for a reasonable cost ($4-5k) compared to, say, putting a Blackwell video card in your Windows desktop (run you about $9k).
Is it possible to hook one of these up to my Windows PC and use it to run Topaz Starlight models? Would I even see a speed improvement? If so, how would that compare to one of the more expensive Blackwell video cards?
Hi.
Here’s the list of supported Cards
One of the minimum NVIDIA GPU requirements is that the GPU needs the CUDA compute capability of at least 7.0, which means all the legacy GPUs with compute capability smaller than this version are not supported, including GTX 10 series and Quadro P series. Click here to see a list of supported NVIDIA GPUs.
(3) At least 10 GB of VRAM is required for NVIDIA GPUs, but we strongly recommend a GPU with 16 GB of VRAM or more for optimal performance and quality.
Here’s the original link
Hope this helps
Yes and no. What I linked wasn’t a graphics card, but rather an AI processing unit that runs on Linux, with an integrated Nvidia graphics card that has a compute capability of 12.1.
Currently there is not an option to use the Spark with the standard desktop UI app of Topaz Video. Using something like that would require a custom configuration of the app to access and run via that setup and even then I am not sure how well it would do with the models.
Thank you! Can you confirm if the Blackwell 6000 cards increase performance noticeably in Starlight renders, vs say a 5090?
Depending on the original file resolution size and the amount of upscaling, we did see faster processing with the 6000 over a 5090.
For a rough idea test file was 10 seconds log and 24fps:
1920x1080 source 2x - 31 minutes 27 seconds versus 38 minutes 26 seconds.
720x480 source 4x - 22 minutes 52 seconds vs 27 minutes 20 seconds
Excellent thanks!
